Pydantic creator open-sources Monty: a 1ms-startup Python sandbox that ran 10k scripts in 674ms
samuelcolvin · x · 2026-09-25
Samuel Colvin (creator of Pydantic) has released Monty v1, a Python sandbox built for agent code execution that starts in 1 millisecond versus 1.5s for cloud sandboxes. He benchmarked 10,000 sandboxed scripts in 674ms total — a task that would take a cloud sandbox over 3 hours.
Key features:
- Dump and resume: long-lived sessions can be suspended and resumed at any external function call
- Easy host bridging: trivially expose local host/agent functions into the sandbox
- Tiny footprint: workers use as little as 2MB of memory, suited for massive scale
Open source now via PyPI, npm and Crates; a hosted platform is coming and design partners are being recruited. Colvin argues this removes the biggest drawback of letting agents write code.
Related event: Pydantic Creator Open-Sources Monty, a 1ms-Startup Python Sandbox(4 posts)→
More from coding & agent
- Antithesis teaches agents mutation testing to break distributed safety properties — moarbugs · 2026-09-25
- LlamaParse nails the Fed's dot plot: 9 for 9 on GDP medians — llama_index · 2026-09-25
- Microsoft rebuilds Copilot into a super app with persistent Autopilot agents — rohanpaul_ai · 2026-09-25
- Embedding test: RAG ranks the no-refund policy first, showing models still matter — galratner · 2026-09-25
- Jev for PowerShell 0.2.0 turns messy incident text into page/do-not-page decisions — dfinke · 2026-09-25
- Matt Pocock unveils /pr and /retro at AI Engineer Paris to speed up code reviews — mattpocockuk · 2026-09-25